Large Multidimensional Datasets Research Articles

Gcxgclab: An R Package for Two-Dimensional Gas Chromatography Preprocessing and Analysis.

Comprehensive two-dimensional gas chromatography (GC×GC) is an established technique capable of chromatographically separating thousands of analytes in complex matrices. When coupled with highly sensitive detectors such as a high-resolution mass spectrometer, these instruments produce large multidimensional data sets. A prevailing challenge for GC×GC users is efficient data handling and analysis. Although commercial software packages exist for GC×GC data analysis, these platforms are typically optimized for low-throughput qualitative data interpretation utilizing desktop computer systems. Additionally, commercial GC×GC data processing packages offer little flexibility to explore custom or novel data processing concepts. In this work, an open-source R package, called "gcxgclab", was developed as an alternative data processing package for GC×GC users and offers data preprocessing and analysis functions including baseline correction, smoothing, phase shift, peak detection, peak alignment, extracted ion chromatogram, mass spectra extraction, mass defect analysis, and targeted and nontargeted analysis. This package was designed to be customizable and allows for batch data processing on desktops or high-performance computing systems to increase the throughput. gcxgclab is available for free on the Comprehensive R Archive Network (CRAN) at https://cran.r-project.org/web/packages/gcxgclab/index.html.

Multidimensional Data Processing With Bayesian Inference via Structural Block Decomposition.

How to handle large multidimensional datasets, such as hyperspectral images and video information, efficiently and effectively plays a critical role in big-data processing. The characteristics of low-rank tensor decomposition in recent years demonstrate the essentials in describing the tensor rank, which often leads to promising approaches. However, most current tensor decomposition models consider the rank-1 component simply to be the vector outer product, which may not fully capture the correlated spatial information effectively for large-scale and high-order multidimensional datasets. In this article, we develop a new novel tensor decomposition model by extending it to the matrix outer product or called Bhattacharya-Mesner product, to form an effective dataset decomposition. The fundamental idea is to decompose tensors structurally in a compact manner as much as possible while retaining data spatial characteristics in a tractable way. By incorporating the framework of the Bayesian inference, a new tensor decomposition model on the subtle matrix unfolding outer product is established for both tensor completion and robust principal component analysis problems, including hyperspectral image completion and denoising, traffic data imputation, and video background subtraction. Numerical experiments on real-world datasets demonstrate the highly desirable effectiveness of the proposed approach.

Research progress on the application of 16S rRNA gene sequencing and machine learning in forensic microbiome individual identification.

Forensic microbiome research is a field with a wide range of applications and a number of protocols have been developed for its use in this area of research. As individuals host radically different microbiota, the human microbiome is expected to become a new biomarker for forensic identification. To achieve an effective use of this procedure an understanding of factors which can alter the human microbiome and determinations of stable and changing elements will be critical in selecting appropriate targets for investigation. The 16S rRNA gene, which is notable for its conservation and specificity, represents a potentially ideal marker for forensic microbiome identification. Gene sequencing involving 16S rRNA is currently the method of choice for use in investigating microbiomes. While the sequencing involved with microbiome determinations can generate large multi-dimensional datasets that can be difficult to analyze and interpret, machine learning methods can be useful in surmounting this analytical challenge. In this review, we describe the research methods and related sequencing technologies currently available for application of 16S rRNA gene sequencing and machine learning in the field of forensic identification. In addition, we assess the potential value of 16S rRNA and machine learning in forensic microbiome science.

AIDA: Analytic isolation and distance-based anomaly detection algorithm

Many unsupervised anomaly detection algorithms rely on the concept of nearest neighbours to compute the anomaly scores. Such algorithms are popular because there are no assumptions about the data, making them a robust choice for unstructured datasets. However, the number (k) of nearest neighbours, which critically affects the model performance, cannot be tuned in an unsupervised setting. Hence, we propose the new and parameter-free Analytic Isolation and Distance-based Anomaly (AIDA) detection algorithm, that combines the metrics of distance with isolation. Based on AIDA, we also introduce the Tempered Isolation-based eXplanation (TIX) algorithm, which identifies the most relevant features characterizing an outlier, even in large multi-dimensional datasets, improving the overall explainability of the detection mechanism. Both AIDA and TIX are thoroughly tested and compared with state-of-the-art alternatives, proving to be useful additions to the existing set of tools in anomaly detection.

Computational Flow Cytometry Provides Accurate and Sensitive Measurable Residual Disease Assessment in Chronic Lymphocytic Leukemia

Measurable residual disease (MRD) is a highly sensitive reflection of disease burden that correlates with time-to-event outcomes in chronic lymphocytic leukemia (CLL). Undetectable MRD at the end of fixed-duration treatment is an independent prognostic marker and associated with favourable progression-free and overall survival. Muti-colour flow cytometry (FC) is the most common method of MRD assessment and often performed using a harmonized protocol established by the European Research Initiative on CLL. Conventional FC analysis involves sequential manual gating strategies performed on 2-dimensional scatter plots with varying axis parameters. While straightforward with a small number of parameters, this process becomes too complex and time-consuming with state-of-the-art mass flow cytometers where up to 40 markers are simultaneously measured and large multidimensional datasets are generated. Over the past decade, a variety of computational FC algorithms have been developed to automate and assist in tasks such as population identification, enumeration, and classification. These mathematical approaches are based on dimension reduction and clustering methods that require up to several hours and have not been compatible with routine diagnostic workflows until recently. Flow-Self Organizing Maps (FlowSOM) (Van Gassen et al. Cytometry A 2015) is an unsupervised clustering algorithm (R package from Bioconductor) with fast analysis times and multidimensional visualization. Implementation of FlowSOM in a diagnostic laboratory offers several advantages including lower technical variability, reduced bias, and improved efficiency. We developed a computational pipeline to assess CLL MRD using FlowSOM. CLL flow cytometry standard (FCS) files were identified from the Peter MacCallum Cancer Centre and included cases with typical and atypical immunophenotypes from peripheral blood (PB) or bone marrow (BM). Atypical CLL was defined as cases with different expression of one or more antigens from the typical immunophenotype. The diagnostic workflow was developed using a training set (n=25) containing representative samples from MRD negative (n=5), typical CLL (n=5), atypical CLL (n=10), and healthy bone marrow (n=5) cases. Computational accuracy was evaluated against expert human analysis in a validation cohort (n=293) consisting of PB typical (n=139), PB atypical (n=91), BM typical (n=30), and BM atypical CLL (n=33) samples. FCS files were preprocessed (PeacoQC, FlowCore R packages) to remove low-quality events, apply compensation, and transform the data prior to FlowSOM analysis. Two-level hierarchical clustering was performed to achieve maximum separation between normal and MRD events (level 1: lymphocytes, level 2: B cells). The training set was used to generate a diagnostic map of 225 clusters containing the full breadth of immature and mature B cells along with disease immunophenotypes (Figure 1). Clusters were verified by manual inspection in Kaluza software. Computational MRD was then determined by applying the FlowSOM map to validation samples and counting events within the disease clusters. Computational MRD demonstrated excellent correlation with expert human analysis. Linear regression R2 was 0.96, 0.79, 0.96, and 0.91 for PB typical, PB atypical, BM typical, and BM atypical CLL cohorts, respectively (Figure 2). Positive or negative MRD was highly accurate using a threshold of 0.01% for MRD. Computational analysis demonstrated high sensitivity (99%, 99%, 100%, 90%), specificity (92%, 92%, 100%, 85%), and concordance (95%, 98%, 100%, 88%) across all four validation groups (PB typical, PB atypical, BM typical and BM atypical CLL, respectively). The kappa coefficient was 0.92 across the entire validation cohort. CLL MRD assessment in BM samples can be difficult due to overlapping immunophenotypes particularly between disease and transitional B cells. We subsequently included an additional BM cohort consisting of only MRD negative samples to further assess detectable/undetectable accuracy. Samples were highly concordant (94%) except for two cases with low disease signal (0.010% and 0.012%). In summary, FlowSOM computational CLL MRD is highly accurate and offers the potential for an automated workflow. This method can be adapted to other haematological malignancies including acute leukemia and lymphoproliferative disorders. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Unraveling the complexities of urban fluvial flood hydraulics through AI

As urbanization increases across the globe, urban flooding is an ever-pressing concern. Urban fluvial systems are highly complex, depending on a myriad of interacting variables. Numerous hydraulic models are available for analyzing urban flooding; however, meeting the demand of high spatial extension and finer discretization and solving the physics-based numerical equations are computationally expensive. Computational efforts increase drastically with an increase in model dimension and resolution, preventing current solutions from fully realizing the data revolution. In this research, we demonstrate the effectiveness of artificial intelligence (AI), in particular, machine learning (ML) methods including the emerging deep learning (DL) to quantify urban flooding considering the lower part of Darby Creek, PA, USA. Training datasets comprise multiple geographic and urban hydraulic features (e.g., coordinates, elevation, water depth, flooded locations, discharge, average slope, and the impervious area within the contributing region, downstream distance from stormwater outfalls and dams). ML Classifiers such as logistic regression (LR), decision tree (DT), support vector machine (SVM), and K-nearest neighbors (KNN) are used to identify the flooded locations. A Deep neural network (DNN)-based regression model is used to quantify the water depth. The values of the evaluation matrices indicate satisfactory performance both for the classifiers and DNN model (F-1 scores- 0.975, 0.991, 0.892, and 0.855 for binary classifiers; root mean squared error- 0.027 for DNN regression). In addition, the blocked K-folds Cross Validation (CV) of ML classifiers in detecting flooded locations showed satisfactory performance with the average accuracy of 0.899, which validates the models to generalize to the unseen area. This approach is a significant step towards resolving the complexities of urban fluvial flooding with a large multi-dimensional dataset in a highly computationally efficient manner.

Using data visualization to detect patterns in whole-person health data.

Data visualization techniques are useful for examining large multidimensional data sets. In this exploratory data analysis (EDA) study, we applied a visualization pattern detection and testing process to deidentified data to discover patterns in whole-person health for adults 65 and older. Whole-person health examines a person's environmental, psychosocial, and physical health, as well as their health-related behaviors; and assesses their strengths, challenges, and needs. Strengths are defined as assets and capabilities in the face of short-or long-term stressors. We collected data using a mobile application that delivers a comprehensive whole-person assessment using a simplified version of a standardized instrument, the Omaha System. The visualization pattern detection process is iterative, includes various techniques, and requires visualization literacy. The data visualization techniques applied in this analysis included bubble charts, parallel coordinates line graphs, box plots, and alluvial flow diagrams. We discovered six patterns within the visualizations. We formulated and tested six hypotheses based on these six patterns, and all six hypotheses were supported. Adults 65 and older had more strengths than challenges and more challenges than needs (p < 0.001). Strengths and challenges were negatively correlated (p < 0.001). Unexpectedly, a subset of adults 65 and older who had many, but not all, strengths had significantly more needs (p = 0.04). The use of standardized terminology with its inherent data interrelationships was key to discovering patterns in whole-person health. This methodology may be used in future EDA research using new data sets.

Using Proximity Graph Cut for Fast and Robust Instance-Based Classification in Large Datasets

K-nearest neighbours (kNN) is a very popular instance-based classifier due to its simplicity and good empirical performance. However, large-scale datasets are a big problem for building fast and compact neighbourhood-based classifiers. This work presents the design and implementation of a classification algorithm with index data structures, which would allow us to build fast and scalable solutions for large multidimensional datasets. We propose a novel approach that uses navigable small-world (NSW) proximity graph representation of large-scale datasets. Our approach shows 2–4 times classification speedup for both average and 99th percentile time with asymptotically close classification accuracy compared to the 1-NN method. We observe two orders of magnitude better classification time in cases when method uses swap memory. We show that NSW graph used in our method outperforms other proximity graphs in classification accuracy. Our results suggest that the algorithm can be used in large-scale applications for fast and robust classification, especially when the search index is already constructed for the data.

Machine learning analysis of 137Cs contamination of terrestrial plants after the Fukushima accident using the random forest algorithm

Radioactive contamination of terrestrial plants was extensively investigated and quantitatively modeled after the Fukushima nuclear power plant accident. This phenomenon, which is important for ecosystem functioning and protection of human health, is influenced by multiple factors, including plant species, time after the accident, and climate. Machine learning algorithms such as random forests (RF) have a record of strong performance on large multi-dimensional data sets, but, to our knowledge, combined data on post-Fukushima plant contamination with radionuclides were not yet subjected to a machine learning analysis. Here we performed such analysis on two large published data sets: (1) 137Cs activity concentrations in four common Japanese forest tree species. (2) Plant/soil 137Cs concentration ratios in multiple perennial plant species. The goal was to show the usefulness of machine learning for identifying and quantifying the main trends of 137Cs contamination in terrestrial plants. Each data set was split randomly into training and testing parts, RF was fitted and tuned on the training parts, and its performance was assessed on the testing parts by three metrics: coefficient of determination (R2), root mean squared error, and mean absolute error. Synthetic noise variables and the Boruta algorithm were used in a customized procedure to identify the most important predictor variables, which consistently outperformed random noise. Good agreement between observations and RF predictions (e.g. R2∼0.9 on testing data) was obtained on both data sets. The effects of the most important predictors (e.g. time after the accident, 137Cs land contamination level, and plant species) and interactions between them were quantified by partial dependence plots. These results of machine learning analyses of large data collections can help to complement previous modeling efforts, and to clarify the patterns of 137Cs contamination of plants after the Fukushima accident.

The ImageJ ecosystem: Open-source software for image visualization, processing, and analysis.

For decades, biologists have relied on software to visualize and interpret imaging data. As techniques for acquiring images increase in complexity, resulting in larger multidimensional datasets, imaging software must adapt. ImageJ is an open-source image analysis software platform that has aided researchers with a variety of image analysis applications, driven mainly by engaged and collaborative user and developer communities. The close collaboration between programmers and users has resulted in adaptations to accommodate new challenges in image analysis that address the needs of ImageJ's diverse user base. ImageJ consists of many components, some relevant primarily for developers and a vast collection of user-centric plugins. It is available in many forms, including the widely used Fiji distribution. We refer to this entire ImageJ codebase and community as the ImageJ ecosystem. Here we review the core features of this ecosystem and highlight how ImageJ has responded to imaging technology advancements with new plugins and tools in recent years. These plugins and tools have been developed to address user needs in several areas such as visualization, segmentation, and tracking of biological entities in large, complex datasets. Moreover, new capabilities for deep learning are being added to ImageJ, reflecting a shift in the bioimage analysis community towards exploiting artificial intelligence. These new tools have been facilitated by profound architectural changes to the ImageJ core brought about by the ImageJ2 project. Therefore, we also discuss the contributions of ImageJ2 to enhancing multidimensional image processing and interoperability in the ImageJ ecosystem.

Potentials and caveats of AI in hybrid imaging

State-of-the-art patient management frequently mandates the investigation of both anatomy and physiology of the patients. Hybrid imaging modalities such as the PET/MRI, PET/CT and SPECT/CT have the ability to provide both structural and functional information of the investigated tissues in a single examination. With the introduction of such advanced hardware fusion, new problems arise such as the exceedingly large amount of multi-modality data that requires novel approaches of how to extract a maximum of clinical information from large sets of multi-dimensional imaging data. Artificial intelligence (AI) has emerged as one of the leading technologies that has shown promise in facilitating highly integrative analysis of multi-parametric data. Specifically, the usefulness of AI algorithms in the medical imaging field has been heavily investigated in the realms of (1) image acquisition and reconstruction, (2) post-processing and (3) data mining and modelling. Here, we aim to provide an overview of the challenges encountered in hybrid imaging and discuss how AI algorithms can facilitate potential solutions. In addition, we highlight the pitfalls and challenges in using advanced AI algorithms in the context of hybrid imaging and provide suggestions for building robust AI solutions that enable reproducible and transparent research.

An empirical analysis of graph‐based linear dimensionality reduction techniques

SummaryMany emerging applications such as social networks have prompted remarkable attention in graph data analysis. Graph data is typically high‐dimensional in nature, and dimensionality reduction is critical regarding storage, analysis, and querying of such data efficiently. Although there are many dimensionality reduction methods, it is not clear to what extent the performances of the various dimensionality reduction techniques differ. In this article, we review some of the well‐known linear dimensionality reduction methods and perform an empirical analysis of these approaches using large multidimensional graph datasets. Our results show that in linear unsupervised learning methods, the principal component analysis, singular value decomposition, and neighborhood preserving embedding methods achieve better retrieval data performance than other methods of the statistical information category, dictionary methods, and embedding methods, respectively. Regarding supervised learning methods, the experimental results demonstrate that linear discriminant analysis and partial least squares presented almost similar results.

Fast Pixelated Detectors in Scanning Transmission Electron Microscopy. Part II: Post-Acquisition Data Processing, Visualization, and Structural Characterization.

Fast pixelated detectors incorporating direct electron detection (DED) technology are increasingly being regarded as universal detectors for scanning transmission electron microscopy (STEM), capable of imaging under multiple modes of operation. However, several issues remain around the post-acquisition processing and visualization of the often very large multidimensional STEM datasets produced by them. We discuss these issues and present open source software libraries to enable efficient processing and visualization of such datasets. Throughout, we provide examples of the analysis methodologies presented, utilizing data from a 256 × 256 pixel Medipix3 hybrid DED detector, with a particular focus on the STEM characterization of the structural properties of materials. These include the techniques of virtual detector imaging; higher-order Laue zone analysis; nanobeam electron diffraction; and scanning precession electron diffraction. In the latter, we demonstrate a nanoscale lattice parameter mapping with a fractional precision ≤6 × 10−4 (0.06%).

NeuroPycon: An open-source python toolbox for fast multi-modal and reproducible brain connectivity pipelines

Recent years have witnessed a massive push towards reproducible research in neuroscience. Unfortunately, this endeavor is often challenged by the large diversity of tools used, project-specific custom code and the difficulty to track all user-defined parameters. NeuroPycon is an open-source multi-modal brain data analysis toolkit which provides Python-based template pipelines for advanced multi-processing of MEG, EEG, functional and anatomical MRI data, with a focus on connectivity and graph theoretical analyses. Importantly, it provides shareable parameter files to facilitate replication of all analysis steps. NeuroPycon is based on the NiPype framework which facilitates data analyses by wrapping many commonly-used neuroimaging software tools into a common Python environment. In other words, rather than being a brain imaging software with is own implementation of standard algorithms for brain signal processing, NeuroPycon seamlessly integrates existing packages (coded in python, Matlab or other languages) into a unified python framework. Importantly, thanks to the multi-threaded processing and computational efficiency afforded by NiPype, NeuroPycon provides an easy option for fast parallel processing, which critical when handling large sets of multi-dimensional brain data. Moreover, its flexible design allows users to easily configure analysis pipelines by connecting distinct nodes to each other. Each node can be a Python-wrapped module, a user-defined function or a well-established tool (e.g. MNE-Python for MEG analysis, Radatools for graph theoretical metrics, etc.). Last but not least, the ability to use NeuroPycon parameter files to fully describe any pipeline is an important feature for reproducibility, as they can be shared and used for easy replication by others. The current implementation of NeuroPycon contains two complementary packages: The first, called ephypype, includes pipelines for electrophysiology analysis and a command-line interface for on the fly pipeline creation. Current implementations allow for MEG/EEG data import, pre-processing and cleaning by automatic removal of ocular and cardiac artefacts, in addition to sensor or source-level connectivity analyses. The second package, called graphpype, is designed to investigate functional connectivity via a wide range of graph-theoretical metrics, including modular partitions. The present article describes the philosophy, architecture, and functionalities of the toolkit and provides illustrative examples through interactive notebooks. NeuroPycon is available for download via github (https://github.com/neuropycon) and the two principal packages are documented online (https://neuropycon.github.io/ephypype/index.html, and https://neuropycon.github.io/graphpype/index.html). Future developments include fusion of multi-modal data (eg. MEG and fMRI or intracranial EEG and fMRI). We hope that the release of NeuroPycon will attract many users and new contributors, and facilitate the efforts of our community towards open source tool sharing and development, as well as scientific reproducibility.

Pre-deployment risk factors for PTSD in active-duty personnel\xa0deployed to Afghanistan: a machine-learning approach for analyzing multivariate predictors

Active-duty Army personnel can be exposed to traumatic warzone events and are at increased risk for developing post-traumatic stress disorder (PTSD) compared with the general population. PTSD is associated with high individual and societal costs, but identification of predictive markers to determine deployment readiness and risk mitigation strategies is not well understood. This prospective longitudinal naturalistic cohort study—the Fort Campbell Cohort study—examined the value of using a large multidimensional dataset collected from soldiers prior to deployment to Afghanistan for predicting post-deployment PTSD status. The dataset consisted of polygenic, epigenetic, metabolomic, endocrine, inflammatory and routine clinical lab markers, computerized neurocognitive testing, and symptom self-reports. The analysis was computed on active-duty Army personnel (N = 473) of the 101st Airborne at Fort Campbell, Kentucky. Machine-learning models predicted provisional PTSD diagnosis 90–180 days post deployment (random forest: AUC = 0.78, 95% CI = 0.67–0.89, sensitivity = 0.78, specificity = 0.71; SVM: AUC = 0.88, 95% CI = 0.78–0.98, sensitivity = 0.89, specificity = 0.79) and longitudinal PTSD symptom trajectories identified with latent growth mixture modeling (random forest: AUC = 0.85, 95% CI = 0.75–0.96, sensitivity = 0.88, specificity = 0.69; SVM: AUC = 0.87, 95% CI = 0.79–0.96, sensitivity = 0.80, specificity = 0.85). Among the highest-ranked predictive features were pre-deployment sleep quality, anxiety, depression, sustained attention, and cognitive flexibility. Blood-based biomarkers including metabolites, epigenomic, immune, inflammatory, and liver function markers complemented the most important predictors. The clinical prediction of post-deployment symptom trajectories and provisional PTSD diagnosis based on pre-deployment data achieved high discriminatory power. The predictive models may be used to determine deployment readiness and to determine novel pre-deployment interventions to mitigate the risk for deployment-related PTSD.

Integrating memory-mapping and N-dimensional hash function for fast and efficient grid-based climate data query

ABSTRACT Database systems are pervasive components in the current big data era. However, efficiently managing and querying grid-based or array-based multidimensional climate data are still beyond the capabilities of most databases. The mismatch between the array data model and relational data model limited the performance to query multidimensional data in a traditional database when data volume hits a cap. Even a trivial data retrieval on large multidimensional datasets in a relational database is time-consuming and requires enormous storage space. Given the scientific interests and application demands on time-sensitive spatiotemporal data query and analysis, there is an urgent need for efficient data storage and fast data retrieval solutions on large multidimensional datasets. In this paper, we introduce a method for multidimensional data storing and accessing, which includes a new hash function algorithm that works on a unified data storage structure and couples with the memory-mapping technology. A prototype database library, LotDB developed as an implementation, is described in this paper, which shows promising results on data query performance compared with SciDB, MongoDB, and PostgreSQL.

Pre-processing visualization of hyperspectral fluorescent data with Spectrally Encoded Enhanced Representations

Hyperspectral fluorescence imaging is gaining popularity for it enables multiplexing of spatio-temporal dynamics across scales for molecules, cells and tissues with multiple fluorescent labels. This is made possible by adding the dimension of wavelength to the dataset. The resulting datasets are high in information density and often require lengthy analyses to separate the overlapping fluorescent spectra. Understanding and visualizing these large multi-dimensional datasets during acquisition and pre-processing can be challenging. Here we present Spectrally Encoded Enhanced Representations (SEER), an approach for improved and computationally efficient simultaneous color visualization of multiple spectral components of hyperspectral fluorescence images. Exploiting the mathematical properties of the phasor method, we transform the wavelength space into information-rich color maps for RGB display visualization. We present multiple biological fluorescent samples and highlight SEER’s enhancement of specific and subtle spectral differences, providing a fast, intuitive and mathematical way to interpret hyperspectral images during collection, pre-processing and analysis.

Forensic Prediction using Bias-Variance Tradeoff and Random Forest Algorithms

Every individual host after death has its own altered micro biome configuration. After death, postmortem microorganism communities change to represent the attributes of death. The micro biome act as a many roles in human health, usually done by the exclusive lens of clinical interest. By scouring 5 anatomical areas throughout regular demise exploration from 188 case to predict the Postmortem Interval (PMI), location of death and manner of death, the postmortem micro biomes were collected. The micro biome sequencing are not easy to analyze and interpret because it produces large multidimensional dataset. To overcome the analytical challenge Machine learning method can be used. The two supervised machine learning methods employed here are Random Forest and Bias-Variance Tradeoff. In training datasets, Random forest algorithm is applied. This algorithm makes predictions by choosing the most voted node from each decision tree as the output. The output is checked for any bias variance error, by the Bias-Variance Tradeoff algorithm in order to help the supervised learning algorithm to perform generalization beyond the training datasets. To obtain a prediction that is best fitted and accurate, these two algorithms are chosen for learning.

INNV-15. CLINICAL DATA THAT MATTERS: A DISTILLATION OF NEURO-ONCOLOGY CLINICAL TRIAL INCLUSION CRITERIA USING MACHINE LEARNING

Abstract INTRODUCTION Neuro-oncologic conditions have dismal outcomes, ineffective treatments, poor access to clinical trials, and variability in care. Clinical trials do not capture a patient’s complete journey and are restricted to select populations. ‘Real-world-evidence’ (RWE) attempts to inform point of care decisions through routine collection of data with a clinical-trial-like rigor. RWE complements existing knowledge through broad patient participation, collection throughout disease course, and creation of large multidimensional datasets “knowledge network of disease” 1,2. RWE implementation is hindered by unstructured data, uncertainty of relevant features, and semantic heterogeneity. Clinical attributes were selected from trial inclusion criteria and prioritized for structuring in clinic notes for abstraction. METHOD We queried Clinicaltrials.gov from 1/1/2018-12/31/2018, refined to North America, recruiting, interventional, and adult. Meningioma, pituitary, glioblastoma, astrocytoma, oligodendroglioma, and ependymoma were chosen based on incidence3. Lymphoma and nerve sheath tumors were omitted. “Brain tumor” and “glioma” were added. ‘K-nearest-neighbor’ tokenization parsed inclusion criteria4. Document term matrix (n-gram) converted text to vectors5. A generative probabilistic model using ‘Latent Dirichlet Allocation’ plotted words into 10 clusters6. Hierarchal clustering was used to compare histology with terms. RESULTS 401 trials parsed into 3676 statements and 4008 keywords. 10 clusters of terms were similarly distributed amongst histologies, suggesting generalizability across tumor types. Cluster revealed 8 categories: 1) Time: enrollment; 2) Performance status: KPS; 3) Testing: mutations, upper limit of normal, routine hematologic laboratory assays; 4) Imaging: extent of surgery; 5) Pregnancy/childbearing; 6) Tumor grade; 7) Treatment history: recurrence, chemotherapy, radiation, time; 8) Informed consent CONCLUSIONS Dissecting the compendium of clinical trials using machine learning can identify general parameters for trial enrollment to guide RWE clinical collection. Using practical definitions of the most germane trial data, specific information can be sought after and defined to improve research quality, maximize research yields and improve patient care whilst minimizing wasted research and clinical endeavors.

Cinematic Visualization of Multiresolution Data: Ytini for Adaptive Mesh Refinement in Houdini

Abstract We have entered the era of large multidimensional data sets represented by increasingly complex data structures. Current tools for scientific visualization are not optimized to efficiently and intuitively create cinematic production quality, time-evolving representations of numerical data for broad impact science communication via film, media, or journalism. To present such data in a cinematic environment, it is advantageous to develop methods that integrate these complex data structures into industry-standard visual effects software packages, which provide a myriad of control features otherwise unavailable in traditional scientific visualization software. In this paper, we present the general methodology for the import and visualization of nested multiresolution data sets into commercially available visual effects software. We further provide a specific example of importing adaptive mesh refinement data into the software Houdini. This paper builds on our previous work, which describes a method for using Houdini to visualize uniform Cartesian data sets. We summarize a tutorial available on the website www.ytini.com, which includes sample data downloads, Python code, and various other resources to simplify the process of importing and rendering multiresolution data.